ChatIA: the institutional generative AI platform of the Public Prosecutor's Office of the Province of Buenos Aires

The problem

The Public Prosecutor's Office of the Province of Buenos Aires operates at a significant scale. The province covers an area of 305,907 km² (8.2% of Argentina's national territory) and has a population of 17.5 million inhabitants. The office comprises more than 8,700 public officials and participates in approximately 1 million criminal proceedings per year.

The main challenge is to close the gap between the speed of technological advancement and the pace at which people and institutions can adapt. Generative AI holds significant potential to strengthen judicial work and improve the delivery of justice services. That potential, however, is only realised if its use is safe and responsible, with protection of sensitive information, traceability of processes, and human oversight of decisions that affect rights. The challenge, then, is to harness that power without giving up institutional control.

The solution

ChatIA is the institutional generative AI platform of the Public Prosecutor's Office of the Province of Buenos Aires. This project began in 2025 with the goal of promoting the responsible adoption of AI tools within the justice system. ChatIA was designed and developed by the Public Prosecutor's Office, through its Secretariat of Innovation and Digital Experience and its Undersecretariat of Informatics, to support judicial officers in performing their duties, improve the efficiency of the justice service, and ensure data protection, traceability, and human oversight of AI use.

A centralised and controlled platform is what makes it possible to harness the potential of generative AI while protecting sensitive information and keeping human judgement at the centre of every decision that affects rights, through an infrastructure that guarantees confidentiality, traceability, and continuous human oversight.

The platform integrates multiple AI models within a centralised architecture that includes user authentication, role-based access control (where different staff have access to different functions depending on their role), interaction monitoring, and usage auditing. Priority is given to using locally hosted models to protect institutional information, while maintaining the option to use external models under controlled conditions.

The implementation has been organised in progressive stages.

The first stage involved a pilot test and experimentation with ChatIA on tasks that support judicial staff, such as drafting, editing, and summarising documents, as well as facilitating the search for legal opinions and case law. The goal of the pilot was to assess the tool's relevance, security, and usefulness in controlled environments.

Once its reliability and accuracy were verified in the pilot, the project moved into a second stage that incorporated new cases involving ongoing proceedings. In addition, after identifying the need, a library of high-performing prompts is currently being developed. This helps staff achieve consistent, high-quality results on the platform by providing tested and refined ways to phrase requests to the AI.

Also among the use cases, a module was designed to contrast the classification of an offence with the account of the facts under investigation, analysing their level of coherence and consistency. This seeks to strengthen the quality, consistency, and traceability of recording and analysis processes, thereby contributing to more informed decision-making, without in any way replacing the judgement or legal responsibility of judicial operators. At the same time, new use cases are being developed that will positively impact judicial staff's work.

The platform is currently in the pilot testing phase, involving representatives from different areas of the Public Prosecutor's Office.

The impact

1. Pilot demonstrated the potential of generative AI within a controlled judicial environment
The pilot test has demonstrated the enormous potential of generative AI, when adopted securely, to strengthen the functions of the Public Prosecutor's Office. The expansion of the system is expected to generate significant improvements in productivity, in the quality of the documents produced, and in the safer use of AI within the organisation.

2. Currently available to pilot users, with expansion to 1,000 planned
Currently, ChatIA is available to a group of users from different areas of the organisation. In the next stage, it is expected to be made available to 1,000 users. Given that the office has more than 8,700 officials handling approximately one million criminal proceedings per year, even modest efficiency gains per user could have a substantial cumulative effect across the institution.

3. Future plans
In an upcoming stage, the project envisions developing AI agents for assisted legal validation. A pilot test has already been carried out with one such agent, which supports the IT team in its daily work by enabling queries of code repositories that were previously indexed and organised using vectorisation techniques. Vectorisation is a method for converting text into a numerical format that AI systems can search and compare efficiently, enabling quick code discovery across large collections.

The integration of generative AI into the virtual assistant "Vicky" is also planned. Implemented in 2019, Vicky was the first system of its kind in the country's judiciary, and it provides ongoing information about people's rights and the functions of the Public Prosecutor's Office, with the aim of improving the quality of service and offering more accessible and effective assistance. Integrating generative AI would enable Vicky to handle a wider range of queries and respond in a more conversational, context-aware way.

Lessons learned

• Adopting generative AI in the justice sector requires technology, governance, and organisational change to advance together. The main lesson is that adopting generative AI in the justice sector is not a technological problem but a combination of three dimensions that must advance together: technology, institutional governance, and organisational change.
• Gradual implementation through narrow, verifiable use cases worked best. What worked best was a gradual implementation, through narrow, verifiable use cases rather than large-scale rollouts, backed by the institutional leadership of the Attorney General and by continuous monitoring that allowed the team to adjust course along the way. It was also decisive to build on what constitutes the office's greatest strength: valuable, context-specific data and deep institutional knowledge of the Public Prosecutor's Office's processes.
• Start with areas where institutional data and domain knowledge provide an advantage. If starting over, the team would bring that focus in from day one: prioritising the areas where their data and domain knowledge give them a real advantage, rather than general-purpose applications.

ChatIA was developed in accordance with institutional rules approved by the Public Prosecutor's Office to promote the responsible use of generative AI in the judicial sphere. These rules recognise AI as a support tool for judicial officers and are grounded in the principles of personal data protection, information security, transparency, accountability, and human oversight of its outputs. Together, they provide a governance framework that allows the institution to harness this innovation while safeguarding individuals' rights and the guarantees inherent to the administration of justice.

To support this process, a collaboration agreement was signed with Microsoft, the first of its kind in a judicial branch in Latin America, aimed at strengthening digital literacy, technical skills, and productivity across the judicial sector.